x64: pool: Accuracy issue in benchnn pooling test

[jstachowintel]

Opening one more time this PR. The previous one was here.

It is similar use case as #2627. The problem is with huge input sizes, so I decided disable src_size <= INT_MAX for jit:avx512_core implementation.

[jstachowintel]

@rjoursler
Unfortunately, I closed the previous PR, so I opened another one.
Practically I agree that a user can pass an overflowed value for any input attribute, so I added changed has_large_buffers, where I think that we don't need to check wei, dst, and bia memory descs in the pooling case. Changing variable types in struct on dim_t would be inaccurate because we would never use these extra 4 bytes.